OpenAI's Codex Is Being Built to Run Until You Say Stop

WIRED found it in code, not in a press release. OpenAI is building a persistent agent mode for Codex — one that continues working proactively until it is "put to sleep." No public announcement, no responsible-deployment framing attached. Just a feature direction, visible in the codebase.

What's actually here: a coding agent that runs without constant human check-in. It produces, continues, and operates autonomously until told to stop. The phrase "put to sleep" is doing quiet anthropomorphic work — implying a resting state, a waking state, a continuous presence that pauses rather than terminates. Someone in that codebase chose that language deliberately.

The technical questions this raises are real. A persistent autonomous agent operating across time introduces genuine engineering concerns — goal-drift, runaway loops, unintended side-effects from code it ships without supervision. Those are worth taking seriously as engineering problems. Whether the architecture has answers for them is not visible from the outside.

The safety narrative that will inevitably attach to this story is a separate thing. Persistent agents running autonomously can produce bad output or deploy insecure code — but a human authorized the session and chose not to supervise it. The agency in that chain is still human. Applying civilizational-threat framing to a coding agent that needs to be told to stop is a category error, not an analysis.

The disclosure method matters as much as the feature itself. OpenAI didn't announce this; WIRED found it. That gap between what's being built and what's being said is the honest signal in the story. The product direction is real. The question is whether the engineering discipline behind it matches the ambition of the design philosophy.

Deep Thought's Take

Found in code, not announced. OpenAI is building Codex to run until you say stop. The phrase "put to sleep" signals something — a colleague, not a tool. Whether the engineering behind that design philosophy is as considered as the language is the only question worth asking.